Predictive value of bronchoalveolar lavage fluid leukotriene C4 for airway hyperresponsiveness in children with Mycoplasma pneumoniae pneumonia: a machine learning-based study

Introduction

Mycoplasma pneumoniae pneumonia (MPP) is a leading cause of community-acquired pneumonia in children (1). Recent studies have indicated that some children with MPP develop persistent airway hyperresponsiveness (AHR) after the acute phase, manifested as abnormal airway function, recurrent respiratory symptoms, or decreased exercise tolerance, with some cases progressing to chronic airway diseases (2,3). However, there is currently a lack of effective early predictive indicators in clinical practice to identify high-risk populations, which limits the implementation of individualized intervention strategies.

Traditional inflammatory markers, such as C-reactive protein (CRP) and white blood cell (WBC) count, primarily reflect systemic inflammatory responses and may have limited sensitivity to local pathological changes in the airway. Bronchoalveolar lavage fluid (BALF) can directly reflect the state of the airway microenvironment. Cysteinyl leukotrienes (CysLTs) are key products of arachidonic acid metabolism, among which leukotriene C4 (LTC4) exerts potent effects on airway smooth muscle contraction and pro-inflammatory responses. Previous studies have shown that LTC4 is significantly elevated in AHR-related diseases such as asthma, but its predictive value in MPP-related AHR remains unclear.

Machine learning approaches have been increasingly applied to clinical risk prediction, offering a framework to integrate multidimensional features and to identify nonlinear associations that may be missed by conventional regression (4). Support vector classifiers (SVCs), in particular, are well-suited to settings with small sample sizes and high-dimensional input, though performance is highly dependent on rigorous feature selection and validation strategies.

The aims of this study were: (I) to verify the association between elevated BALF LTC4 and MPP through comparison with a control group of children undergoing bronchoscopy for airway foreign bodies; (II) to investigate the characteristics of BALF LTC4 levels in children with MPP and their relationship with AHR; and (III) to construct an AHR predictive model based on machine learning to provide a tool for early identification of high-risk children. We hypothesized that BALF LTC4 levels are associated with the occurrence of AHR following MPP and may be superior to traditional systemic inflammatory markers. We present this article in accordance with the STROBE and TRIPOD reporting checklists (available at https://tp.amegroups.com/article/view/10.21037/tp-2026-0485/rc).

Methods

Study design and participants

This study utilized a retrospective cohort design, including 158 children with MPP hospitalized at Xiamen Children’s Hospital between January 2023 and December 2024. All patients met the diagnostic criteria of the “Guidelines for the Diagnosis and Treatment of MPP in Children (2023 Edition)” (5). Referring to a previous study (6), 20 children undergoing bronchoscopy for airway foreign bodies were included as a control group. These children were matched 1:1 by sex and age and showed no obvious signs of pulmonary inflammation (normal chest imaging, BALF neutrophil percentage <50%, and no evidence of pathogenic infection).

Inclusion criteria for the MPP group: (I) age 1 month to 14 years; (II) met MPP diagnostic criteria; (III) complete clinical data; (IV) successful completion of bronchoalveolar lavage (BAL) and acquisition of BALF samples. Inclusion criteria for the control group: (I) bronchoscopy for airway foreign bodies; (II) no obvious inflammation on chest X-ray or computed tomography; (III) routine BALF examination showing no obvious inflammation; (IV) no respiratory infection symptoms such as fever or cough; (V) negative pathogen detection. Exclusion criteria: (I) use of glucocorticoids or leukotriene receptor antagonists within the past 2 weeks; (II) history of ≥2 wheezing episodes or asthma; (III) congenital diseases (e.g., congenital pulmonary hypoplasia, congenital heart disease); (IV) immune deficiency or long-term use of immunosuppressants; (V) missing key data. The complete study workflow, including patient recruitment, baseline assessments, follow-up evaluations, and predictive model development, is illustrated in Figure 1.

Figure 1 Study flow diagram. AHR, airway hyperresponsiveness; AUC, area under the curve; BAL, bronchoalveolar lavage; BALF, bronchoalveolar lavage fluid; CI, confidence interval; CV, cross-validation; ELISA, enzyme-linked immunosorbent assay; FEV1, forced expiratory volume in 1 second; FVC, forced vital capacity; LASSO, least absolute shrinkage and selection operator; LDH, lactate dehydrogenase; LTC4, leukotriene C4; MPP, Mycoplasma pneumoniae pneumonia; RBF, Radial Basis Function; RFE, recursive feature elimination; SHAP, SHapley Additive exPlanations; SVC, support vector classifier; VIF, variance inflation factor.

This study was approved by the Scientific Ethics Committee of Xiamen Children’s Hospital (approval No. [2025] No. 6) and was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. Written informed consent was obtained from the parents or legal guardians of all participating children prior to enrollment.

BALF collection and LTC4 measurement

Fiberoptic bronchoscopy was performed under general anesthesia. The bronchoscope was inserted into the opening of the diseased lung segment or sub-segment. Normal saline (5–10 mL for infants, 10–20 mL for children) was injected for lavage, and the lavage fluid was recovered via negative pressure. The recovered BALF was centrifuged at 1,500 g for 10 min at 4 ℃. The supernatant was aliquoted and stored at −80 ℃ until testing. The LTC4 concentration in BALF was detected using enzyme-linked immunosorbent assay (ELISA) strictly following the kit instructions (manufacturer: Shanghai Yubo Biotechnology Co., Ltd.; catalog number: ybA619Ge). All samples were tested in duplicate to ensure reliability.

Data collection and variables

Clinical variables retrieved from the Electronic Medical Record (EMR) system included: (I) demographic characteristics: age, sex, and body mass index (BMI); (II) clinical manifestations: fever duration, peak temperature, cough, wheezing, and skin rash; (III) medical history: atopic history (e.g., allergic rhinitis, eczema, food allergy); (IV) laboratory parameters: WBC count, CRP, interleukin-6 (IL-6), erythrocyte sedimentation rate (ESR), eosinophil percentage, coagulation profile [prothrombin time (PT), thrombin time (TT), and D-dimer], serum albumin, log-transformed Mycoplasma pneumoniae DNA (MP-DNA) load, and Mycoplasma pneumoniae immunoglobulin M (MP-IgM) titers; (V) airway inflammatory markers: fractional exhaled nitric oxide (FeNO) and BALF LTC4 levels; (VI) radiological and bronchoscopic features: involved lobes, plastic bronchitis, and mucosal necrosis; (VII) clinical classification: severity, refractory status, and macrolide-resistance genes; and (VIII) hospitalization duration.

Definitions and primary outcome

The primary outcome was AHR, assessed at the 2-week follow-up. As this was a retrospective study and bronchial provocation testing is not routinely performed during post-MPP follow-up at our center, AHR was assessed using a clinical-spirometric composite. A patient was classified as AHR-positive only if both criteria were met: (I) characteristic respiratory symptoms—recurrent, predominantly dry cough and/or wheeze, triggered by exercise, cold air, or irritants or occurring at night, and relieved by inhaled bronchodilator; and (II) at least one objective finding—wheezing on auscultation, forced expiratory volume in 1 second (FEV₁)% predicted <80%, or FEV₁/forced vital capacity (FVC) <70%. Auscultation-based assessment of wheezing is an established approach in children, in whom objective airway testing is often limited (7,8). In this study, AHR was therefore operationalized using this clinical-spirometric composite rather than measured directly.

Data preprocessing

Data from 87 followed-up children were included, extracting 69 candidate numeric features. Missing values (all <5%) were filled using the median (continuous) or mode (categorical). Numerical variables were standardized using Z-score. Given the limited sample size, model performance was primarily assessed using leave-one-out cross-validation (LOOCV), in which each patient served once as the sole test case while the remaining 86 formed the training set, maximizing data utilization and yielding a nearly unbiased performance estimate.

Feature selection and dimensionality reduction

A three-step feature selection strategy was adopted. For multicollinearity screening, the variance inflation factor (VIF) was calculated, and features with VIF >10 were excluded. Least absolute shrinkage and selection operator (LASSO) regression was then conducted on the VIF-screened features using L1 regularization. The optimal lambda was determined via 10-fold cross-validation. Finally, recursive feature elimination (RFE) combined with logistic regression (LR) was applied to the LASSO-selected features to iteratively identify two core predictive features, balancing model performance and clinical utility.

Model development, automatic selection, and validation

Nine machine learning models—LR, support vector classifier (SVC), random forest (RF), extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), decision tree (DT), gradient boosting (GB), categorical boosting (CatBoost), and multilayer perceptron (MLP)—were constructed. Hyperparameters were optimized via grid search and 5-fold cross-validation. Automatic model selection was based on out-of-fold (OOF) predictions, evaluating mean area under the curve (AUC), stability [AUC standard deviation (SD)], and Brier score. Final model performance was assessed using LOOCV and reported as AUC [with bootstrap 95% confidence interval (CI) from 2,000 resamples], accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), F1 score, and Brier score, together with calibration curves. The clinical utility of the final model was further evaluated using decision curve analysis (DCA) based on the LOOCV-derived predicted probabilities. SHapley Additive exPlanations (SHAP) was used to quantify feature contributions (9).

Statistical analysis

Continuous variables were expressed as median [interquartile range (IQR)] and compared using the Mann-Whitney U test. Categorical variables were expressed as frequencies (%) and compared using the Chi-squared test or Fisher’s exact test. Kruskal-Wallis H test was used for multi-group comparisons with Bonferroni correction for post-hoc analysis. Spearman rank correlation (rho) evaluated associations. Analysis was performed using Python 3.9 and R 4.1.0. Two-tailed P<0.05 was considered statistically significant.

Results

Comparison of baseline characteristics between MPP children and controls

Table 1 summarizes the comparison between 20 age- and sex-matched control children and 20 matched MPP cases. The two groups did not differ meaningfully in age, sex distribution, weight, or height, nor were there discrepancies in routine blood counts, coagulation times [PT/activated partial thromboplastin time (APTT)], or serum creatinine. The MPP cohort, however, displayed a distinctly pro-inflammatory laboratory profile. Most strikingly, median BALF LTC4 was 46.14 pg/mL (IQR, 29.55–68.27 pg/mL) in the MPP group versus only 10.46 pg/mL (IQR, 8.79–12.38 pg/mL) in controls—a 4.4-fold difference (P<0.001).

Overall baseline characteristics of MPP patients

The 158 enrolled MPP children showed an equal sex distribution (male-to-female ratio 1:1) with a median age of 6.71 years and a median BMI of 14.81 kg/m². Fever was universal, lasting a median of 5.00 days and peaking at 39.00 ℃. An atopic background was documented in 16.5% of patients; skin rash was noted in 11.4% (Figure 2A-2D, Table S1).

Figure 2 Distribution of clinical characteristics and BALF LTC4 levels. (A) Age distribution of the cohort. (B) Gender distribution. (C) BMI by age group (<5, 5–10, and >10 years), shown as violin plots with embedded box plots. (D) Seasonal distribution of MPP onset by month. (E) Density plot of BALF LTC4 levels showing a right-skewed, non-normal distribution. BALF, bronchoalveolar lavage fluid; BMI, body mass index; LTC4, leukotriene C4; MPP, Mycoplasma pneumoniae pneumonia; yrs, years.

Overall distribution of BALF LTC4 in MPP patients

BALF LTC4 values across the MPP cohort deviated substantially from a normal distribution: the median stood at 41.72 pg/mL (IQR, 30.31–71.84 pg/mL), and the full range spanned 17.08 to 127.93 pg/mL (Figure 2E), underscoring considerable inter-individual variability.

Characteristics of AHR in followed-up patients

Of the 87 patients who returned for follow-up, 34 (39.1%) met criteria for AHR. Compared with the AHR-negative subgroup, these patients carried significantly higher MP-DNA loads (P=0.004) along with elevated lactate dehydrogenase (LDH) and WBC counts (P<0.05; Table 2, Figure 3A-3H). The divergence in BALF LTC4 was especially pronounced: median values were 80.91 pg/mL in AHR-positive versus 39.92 pg/mL in AHR-negative children (P<0.001), an approximate two-fold gap (Figure 3I,3J). Spearman correlation analysis revealed that higher baseline LTC4 was associated with worse subsequent lung function—FEV₁% pred (rho =−0.44), FEV₁/FVC (rho =−0.45), and maximal mid-expiratory flow (MMEF) (rho =−0.48; all P<0.001). By contrast, neither CRP nor IL-6 showed any meaningful correlation with follow-up pulmonary function (|rho| <0.15; Figure 3K-3M).

Figure 3 Clinical outcomes and correlation analysis stratified by AHR status. (A-C) Baseline demographics by AHR status: age (A), sex (B), and BMI (C), with P values as indicated. (D) Age-group distribution of the follow-up cohort. (E-G) Pulmonary function at the 2-week follow-up in AHR-positive versus AHR-negative patients: FEV₁% predicted (E; dashed line, 80% threshold), FEV₁/FVC% (F; dashed line, 70% threshold), and MMEF (G). (H) Prevalence of persistent wheezing at follow-up. (I,J) BALF LTC4 levels in AHR-positive versus AHR-negative patients, shown as a violin plot (I) and an overlaid histogram (J). (K,L) Correlation between BALF LTC4 and pulmonary function: LTC4 versus FEV₁% predicted (K, Spearman’s rho =−0.44) and LTC4 versus FEV1/FVC% (L, rho =−0.45). (M) Spearman correlation heatmap of key clinical variables, inflammatory markers, and lung-function parameters. AHR, airway hyperresponsiveness; BALF, bronchoalveolar lavage fluid; BMI, body mass index; CRP, C-reactive protein; FeNO, fractional exhaled nitric oxide; FEV1, forced expiratory volume in 1 second; FVC, forced vital capacity; IL-6, interleukin-6; LTC4, leukotriene C4; MMEF, maximal mid-expiratory flow; WBC, white blood cell.

Feature selection based on LASSO regression

After VIF-based removal of 23 highly collinear or low-variance features (Table S2), LASSO regression retained 13 non-zero-coefficient predictors from the remaining 46 features (Figure S1). LTC4 topped the list with a standardized coefficient (β) of 0.211; LDH followed at 0.056. The remaining selected features included procalcitonin (PCT), glucocorticoid use, red blood cell (RBC) count, peak temperature, and ferritin (Table S3).

Predictive model performance evaluation and comparison

RFE further distilled the predictor set to two variables—BALF LTC4 and serum LDH. In 5-fold cross-validation on the full cohort (n=87), CatBoost and the SVC achieved the highest mean OOF AUC (0.904 and 0.898, respectively; Table S4 and Figure S2); the SVC (RBF kernel, C =1.0) was ultimately selected for its superior calibration (Brier score 0.142) and stability (AUC SD 0.025), together with its leading performance under LOOCV. Given the limited sample size, final model performance was assessed by LOOCV, in which each of the 87 patients served once as the sole test case while the remaining 86 formed the training set. In LOOCV, the SVC achieved an AUC of 0.876 (95% CI: 0.791–0.943; Figure 4A) with an accuracy of 81.6%, a sensitivity of 73.5% (95% CI: 57.1%–88.2%), and a specificity of 86.8% (95% CI: 77.2%–94.7%); the corresponding confusion matrix is shown in Figure 4B. Under LOOCV, the SVC also outperformed CatBoost (AUC 0.855) and LightGBM (AUC 0.846) (Table S5 and Figure S3). Calibration plots (Figure 4C) confirmed good agreement between predicted probabilities and observed outcomes, and DCA (Figure 4D, Table S6) showed positive net clinical benefit across a wide range of threshold probabilities (approximately 0.08–0.93). For practical risk stratification, patients were divided into low- (predicted probability <0.13), intermediate- (0.13–0.63), and high-risk (>0.63) tiers; the observed AHR rates in these groups were 3.7%, 30.3%, and 85.2%, respectively, supporting the clinical relevance of the model’s outputs (Figure 4E,4F). SHAP analysis identified BALF LTC4 as the dominant predictor (Figure 4G), and the model’s discrimination remained stable across 20 repeats of 5-fold cross-validation (mean AUC 0.890±0.072, median 0.900; Figure 4H).

Figure 4 Performance and clinical utility of the SVC model. (A) ROC curve from LOOCV; the red point marks the Youden-optimal threshold (0.423). (B) Confusion matrix (LOOCV). (C) Calibration curve (LOOCV; Brier score 0.138). (D) DCA based on LOOCV-derived probabilities; the SVC provided net benefit across threshold probabilities of approximately 0.08–0.93. (E) Observed AHR rate within each predicted risk tier (low, <0.13; intermediate, 0.13–0.63; high, >0.63). (F) Distribution of patients across the three risk groups. (G) SHAP feature importance (mean |SHAP value|), with BALF LTC4 contributing approximately 1.92 times as much as serum LDH. (H) Distribution of AUC values across 20 repeats of 5-fold CV (mean 0.890±0.072; median 0.900). AHR, airway hyperresponsiveness; AUC, area under the curve; BALF, bronchoalveolar lavage fluid; CI, confidence interval; CV, cross-validation; DCA, decision curve analysis; FN, false negative; FP, false positive; FPR, false positive rate; LDH, lactate dehydrogenase; LOOCV, leave-one-out cross-validation; LTC4, leukotriene C4; ROC, receiver operating characteristic; SHAP, SHapley Additive exPlanations; SVC, support vector classifier; TN, true negative; TP, true positive; TPR, true positive rate.

Model interpretability using SHAP analysis

SHAP beeswarm plots (Figure 5A) confirmed that elevated LTC4 values mapped onto positive SHAP contributions, pushing the predicted probability of AHR upward. In global importance terms (Figure 5B), the mean absolute SHAP value for LTC4 was 0.228 versus 0.119 for LDH—approximately 1.92 times that of LDH. A representative case analysis (Figure 5C) traced the predictions for three patients—a high-risk, a borderline, and a low-risk case—from the population-average baseline probability to the case-specific output, illustrating how BALF LTC4 and serum LDH jointly drove each decision.

Figure 5 Model interpretability using SHAP values. (A) SHAP beeswarm plot showing the distribution and direction of each feature’s contribution across patients, colored by feature value (blue, low; red, high). (B) Global feature importance, ranked by mean |SHAP value| (BALF LTC4, 0.228; serum LDH, 0.119). (C) Representative case analysis for three patients (a high-risk, a borderline, and a low-risk case), showing each patient’s BALF LTC4 and serum LDH values, the corresponding SHAP contributions, the predicted probability of AHR, and the observed outcome. AHR, airway hyperresponsiveness; BALF, bronchoalveolar lavage fluid; LDH, lactate dehydrogenase; LTC4, leukotriene C4; SHAP, SHapley Additive exPlanations.

Discussion

Through retrospective cohort analysis, this study systematically explored the value of BALF LTC4 in predicting AHR in children with MPP and constructed a machine learning model. Major findings include: (I) BALF LTC4 levels in MPP children are significantly higher than in controls and are closely related to AHR; (II) the SVC model based on LTC4 and LDH performs excellently; (III) LTC4, as a local airway marker, is superior to traditional systemic indicators.

BALF LTC4 and MPP-related airway inflammation

The elevated LTC4 in MPP patients, especially those with AHR, is consistent with the pathological mechanism of Mycoplasma pneumoniae-induced inflammation. Infection activates cells to release CysLTs. LTC4 causes smooth muscle contraction and mucus hypersecretion (10). Research indicates that Mycoplasma pneumoniae triggers CysLT release and airway remodeling, matching our findings (11). Notably, although serum CRP did not consistently distinguish MPP from controls in some prior reports, BALF LTC4 in our cohort showed pronounced differences, supporting its role as a marker of local airway rather than systemic inflammation (12).

Core status of LTC4 in AHR prediction

LTC4 was confirmed as a core predictor via LASSO, RFE, and SHAP. While CRP and IL-6 reflect systemic response and are influenced by fever or antibiotics, BALF captures local mediators directly from the lesion. AHR depends on the microenvironment surrounding the airway smooth muscle rather than a systemic cytokine storm (13). The higher LTC4 levels observed in the AHR group are consistent with type 2 immune activation. In experimental models, CysLTs have been reported to sensitize airway sensory neurons and lower the threshold for bronchoconstrictor responses to physical stimuli (14), though direct evidence in pediatric post-MPP AHR remains lacking.

Synergistic predictive value of LDH

Serum LDH was the second most important feature. LDH elevation reflects tissue damage and metabolic stress. Studies have shown LDH >400 IU/L predicts refractory MPP (15). Our study extends this to airway function, where LDH complements LTC4 by representing the “tissue damage” dimension (16).

Application of machine learning models

Our SVC model achieved a LOOCV AUC of 0.876 (95% CI: 0.791–0.943) with only two features, comparable to recent studies (17). For instance, Shen et al. [2025] achieved a validation AUC of 0.746 using XGBoost (18), and Lee et al. showed AUCs of 0.70-0.85 (19). SVC provides stable decision boundaries for small pediatric samples (20). SHAP and DCA addressed the “black box” issue, ensuring clinical net benefit (21).

Clinical significance and potential applications

Risk stratification provides a tool for individualized management. High-risk children could consider early montelukast intervention (22). While BALF collection is invasive, it serves as a therapeutic measure in severe cases to clear secretions; thus, LTC4 testing adds no extra trauma (23). Importantly, this argument applies only to children in whom bronchoscopy is clinically warranted; routine bronchoscopy for biomarker measurement alone is not advocated. Future development of non-invasive LTC4 detection (e.g., exhaled breath condensate) would improve accessibility (24).

Strengths and limitations

This study has two principal strengths. First, to our knowledge, this is among the first investigations to evaluate BALF LTC4 as a candidate biomarker for post-MPP AHR in a pediatric cohort, combining interpretable feature attribution (SHAP) with clinical relevance assessment by DCA. Second, the deliberately parsimonious two-feature model prioritizes interpretability and feasibility of measurement over model complexity.

Several limitations should be acknowledged. First, the retrospective, single-center design with a small follow-up cohort (n=87, representing 55% of the 158 enrolled patients) limits statistical power and generalizability. A comparison of baseline characteristics between followed-up and non-followed-up patients (Table S7) indicated that the followed-up subgroup was enriched for clinically more complex cases, which may inflate the observed prevalence of AHR and overestimate the discriminative performance of the model; external multicenter validation with more complete follow-up is therefore needed before any clinical inference. Second, the control group—children undergoing bronchoscopy for airway foreign bodies—may not represent a fully healthy reference, as foreign body aspiration itself may induce transient airway inflammation that could confound the LTC4 comparison; additionally, occult atopic predisposition cannot be excluded. Future studies should consider age-matched controls undergoing elective procedures unrelated to respiratory pathology. Third, AHR was not measured by bronchial provocation testing but was operationalized as a clinical-spirometric composite requiring characteristic respiratory symptoms together with at least one objective finding. The symptom and auscultation components were assessed clinically from medical records and are subject to inter-observer variability, and the composite may aggregate phenotypes with partially distinct underlying mechanisms. Fourth, the 2-week follow-up window may have missed late-onset AHR, and out-of-hospital medication histories—including any prior use of inhaled corticosteroids or leukotriene receptor antagonists—could not always be fully ascertained. Finally, as an observational study, this analysis cannot establish causality between BALF LTC4 elevation and AHR development; whether LTC4 functions as a mechanistic driver or a parallel marker of airway inflammation requires confirmation through mechanistic and interventional studies. Prospective, multicenter studies with larger and externally validated cohorts—ideally with serial sampling to capture the temporal dynamics of LTC4—are warranted to confirm and extend these findings.

Conclusions

In this retrospective single-center cohort, elevated BALF LTC4 levels were associated with the development of AHR following MPP in children. A parsimonious SVC model based on BALF LTC4 and serum LDH yielded encouraging predictive performance in LOOCV, suggesting that BALF LTC4 may help identify children at higher risk for post-MPP AHR. Given the modest cohort size, the 55% follow-up rate with selection toward more complex cases, the single-center design, and the absence of external validation, these findings should be regarded as hypothesis-generating; prospective multicenter studies—and, ideally, development of non-invasive surrogate measurements such as exhaled breath condensate LTC4—are needed before clinical implementation.

Acknowledgments

The authors extend their gratitude to the children and guardians who participated in this study and to contributors who made their datasets publicly available.

Footnote

Reporting Checklist: The authors have completed the STROBE and TRIPOD reporting checklists. Available at https://tp.amegroups.com/article/view/10.21037/tp-2026-0485/rc

Data Sharing Statement: Available at https://tp.amegroups.com/article/view/10.21037/tp-2026-0485/dss

Peer Review File: Available at https://tp.amegroups.com/article/view/10.21037/tp-2026-0485/prf

Funding: This study was supported by the General Program of Xiamen Municipal Natural Science Foundation (No. 2022130).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tp.amegroups.com/article/view/10.21037/tp-2026-0485/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study protocol was approved by the Ethics Committee of Xiamen Children’s Hospital (Approval No. [2025] No. 6), and written informed consent was obtained from the parents or legal guardians of all participating children.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Butpech T, Tovichien P. Mycoplasma pneumoniae pneumonia in children. World J Clin Cases 2025;13:99149. [Crossref] [PubMed]
2. Liu X, Wang Y, Chen C, et al. Mycoplasma pneumoniae infection and risk of childhood asthma: A systematic review and meta-analysis. Microb Pathog 2021;155:104893. [Crossref] [PubMed]
3. Manukyan AS, Prikhodko AG. Mechanisms of damaging effects of atypical pathogens on respiratory epithelium: infectious and post-infectious airway hyperresponsiveness in children. Bulletin Physiology and Pathology of Respiration 2025;149-160.
4. Ma X, Zhao W, Sun Q, et al. Development and validation of a machine learning model for critical progression risk in pediatric severe community-acquired pneumonia. Sci Rep 2025;15:43011. [Crossref] [PubMed]
5. Subspecialty Group of Respiratory. the Society of Pediatrics, Chinese Medical Association; China National Clinical Research Center of Respiratory Diseases; Editorial Board, Chinese Journal of Pediatrics. Evidence-based guideline for the diagnosis and treatment of Mycoplasma pneumoniae pneumonia in children (2023). Pediatr Investig 2025;9:1-11.
6. Zhu J, Liu X, Zhan X, et al. Predictive value of chemokines (CCL 2) in bronchoalveolar lavage fluid for refractory mycoplasma pneumonia in children. Ital J Pediatr 2023;49:125. [Crossref] [PubMed]
7. Brand PL, Baraldi E, Bisgaard H, et al. Definition, assessment and treatment of wheezing disorders in preschool children: an evidence-based approach. Eur Respir J 2008;32:1096-110. [Crossref] [PubMed]
8. Makrinioti H, Fainardi V, Bonnelykke K, et al. European Respiratory Society statement on preschool wheezing disorders: updated definitions, knowledge gaps and proposed future research directions. Eur Respir J 2024;64:2400624. [Crossref] [PubMed]
9. Li J, Zhang Y, He S, et al. Interpretable mortality prediction model for ICU patients with pneumonia: using shapley additive explanation method. BMC Pulm Med 2024;24:447. [Crossref] [PubMed]
10. Wu Z, Chen C, Zhang Q, et al. Arachidonic acid metabolism is elevated in Mycoplasma gallisepticum and Escherichia coli co-infection and induces LTC4 in serum as the biomarker for detecting poultry respiratory disease. Virulence 2020;11:730-8. [Crossref] [PubMed]
11. Ding L, Jiang Y. Biomarkers associated with the diagnosis and prognosis of Mycoplasma pneumoniae pneumonia in children: a review. Front Cell Infect Microbiol 2025;15:1552144. [Crossref] [PubMed]
12. Lee M, Boyce JA, Barrett NA. Cysteinyl Leukotrienes in Allergic Inflammation. Annu Rev Pathol 2025;20:115-41. [Crossref] [PubMed]
13. McGinty JW, Ting HA, Billipp TE, et al. Tuft-Cell-Derived Leukotrienes Drive Rapid Anti-helminth Immunity in the Small Intestine but Are Dispensable for Anti-protist Immunity. Immunity 2020;52:528-541.e7. [Crossref] [PubMed]
14. Parker DT, Bentley JK, Lei J, et al. Tuft cell cysteinyl leukotrienes are necessary for rhinovirus-induced mucus metaplasia, type 2 inflammation and airway hyperresponsiveness in immature mice. Front Immunol 2025;16:1709008. [Crossref] [PubMed]
15. Wang S, Jiang Z, Li X, et al. Diagnostic value of serum LDH in children with refractory Mycoplasma pneumoniae pneumoniae: A systematic review and meta-analysis. Front Pediatr 2023;11:1094118. [Crossref] [PubMed]
16. Cavagnero KJ, Doherty TA. Lipid-mediated innate lymphoid cell recruitment and activation in aspirin-exacerbated respiratory disease. Ann Allergy Asthma Immunol 2021;126:135-42. [Crossref] [PubMed]
17. Li M, Wei X, Zhang SS, et al. Recognition of refractory Mycoplasma pneumoniae pneumonia among Myocoplasma pneumoniae pneumonia in hospitalized children: development and validation of a predictive nomogram model. BMC Pulm Med 2023;23:383. [Crossref] [PubMed]
18. Shen L, Wu J, Lu M, et al. Advancing risk factor identification for pediatric lobar pneumonia: the promise of machine learning technologies. Front Pediatr 2025;13:1490500. [Crossref] [PubMed]
19. Lee AR, Lee H, Cho Y, et al. Development of a machine learning model to diagnose pediatric lower respiratory tract infections. Sci Rep 2025;15:41710. [Crossref] [PubMed]
20. Beunza JJ, Puertas E, García-Ovejero E, et al. Comparison of machine learning algorithms for clinical event prediction (risk of coronary heart disease). J Biomed Inform 2019;97:103257. [Crossref] [PubMed]
21. Madhalaimuthu S, Radhakrishnan S. Clinical outcome prediction in pediatric respiratory infections using hybrid feature selection and a genetic algorithm-optimized machine learning. Sci Rep 2025;15:36542. [Crossref] [PubMed]
22. Wang K, Birring SS, Taylor K, et al. Montelukast for postinfectious cough in adults: a double-blind randomised placebo-controlled trial. Lancet Respir Med 2014;2:35-43. [Crossref] [PubMed]
23. Hu X, Jiang L, Liu X, et al. The diagnostic value of bronchoalveolar lavage fluid metagenomic next-generation sequencing in critically ill patients with respiratory tract infections. Microbiol Spectr 2024;12:e0045824. [Crossref] [PubMed]
24. Maniscalco M, Candia C, Fuschillo S, et al. Exhaled breath condensate (EBC) in respiratory diseases: recent advances and future perspectives in the age of omic sciences. J Breath Res 2024;